Greg Stark <gsstark(at)mit(dot)edu> writes:
> There is a simple case that isn't being handled. A straight DISTINCT is
> exactly equivalent to a GROUP BY on all the columns listed. Right now it's
> still doing a Sort+Unique.
True. The DISTINCT logic is so intertwined with ORDER BY that I
couldn't see an easy way to consider a hash-based alternative in the
planner. I think it would require querytree changes propagating all the
way back to the parser to make this feasible at all, and then you'd
still need to think about how to make an honest cost-estimate comparison
of the alternatives (preferably without planning the whole query twice).
Anyone want to dig into it?
regards, tom lane